I’m about 2.5 months in, and still large appliances like washers and even worse the AC (which is one of the primary reasons I wanted a NEST in the first place) aren’t showing up. Instead I seem to be getting a LOT of “refrigerators” and I have one stubborn little “device 1” that turns on for a few seconds now and then.
Seriously, how do you identify a device when sense asks you “is this correct”? I have no way to tell (always) if the AC is running exactly at the time sense claims, we have multiple fridges etc - and I must admit I laughed out loud when I got a “motor detected” device. Like a fan? Or a stepper? Compressor? There are a LOT of motors in electronics … it’s really impossible to confirm/deny if a device found is X. Something should be setup to allow us to verify.
Every time a new device shows up I go straight to settings and tell it to notify me when it turns on and off. The washer, dishwasher, dryer, fridge, stove, etc will show up as multiple devices. For example the appliance light on the fridge will show up as a device. Once it is located, then “merge device”. I have had to merge multiple times. If I am not sure, then I mark it as a guess until I am certain. Once certain, I apply model number to the device for the community. My geothermal HVAC and EV are the stubborn ones. The majority of other devices have arrived. I count around 20 devices in the always on category. Still my always on is under 90 watts so not bad to add up a lot of 1 watt devices. I have about 20 other major devices detected. When they solve my HVAC and EV I will be content to no end.
Looking at the noise issue/learning mode from another angle, would it be worth getting notifications for running appliances at specific times to help train the models?
If I use my toaster in its multiple modes at different times a day, I can use the app to say when I turn it off or on.
Or say when I turn on my dryer which has 10+ modes where sense only sees it running in one mode.
That way the data science team can see when the device is turned on and functioning in different modes and help build more accurate modes than looking at all of the noise without better markers.
Device Integration:
Maybe do a connector to Nest to see when the heat or AC is on to help nail down the bigger appliances and air handlers. This could help with accuracy in the model in the same way the network device detection works. (I won’t use that because I actually VLAN all of my IoT devices into different VLANs from my home network for security.)
I might be misunderstanding. Are you suggesting that you can set up notifications for particular devices so we can get better access to on/off states? It certainly couldn’t hurt, but don’t drive yourself crazy by getting a million notifications a day! We can already comb data for on/off states without notifications (anonymized, of course) and we use that data to build up better models. It’s just a slow, complex process with many, many other inputs.
As for your integration idea, it’s a great one! And yup, very similar to NDI. It’d be a solid method for us to get ground truth about about anything connected into your HVAC system. I can’t give details or timelines, we be are actively looking into the viability of integrations like that
You might setup a beta program or some other user program that allows users that want to contribute additional details to help train the models.
I’m not thinking about knocking myself out trying to give every notification but if I’ve got time I could send a notification when on and then approximately when the device turns off that could help build models to detect devices in other houses more accurately.
Unless my vision of some of this is wrong and detecting devices for each sensor is similar to starting over each time due to different power usage profiles and line noise.
@RyanAtSense yes that’s a feature of the mobile app; but you have to be careful what you ask for. I’m not at home all the time, so getting a ton of notifications isn’t helpful, and very annoying when you’re meeting with customers etc. So I use the history of the device to try to guess - but since it’s in the past, it’s really hard to tell. The garage-door was interesting - since it wasn’t regular but once I could relate it to a few home arrivals/departures a pattern showed up and I got it.
There are quite a few devices you cannot tell when “wake up” and when they don’t. Particular computers can wake up, do processing (more watts) and go back to normal in “random” patterns unless you know EXACTLY what to look for.
Ive heard this is already small part of learning
Process in the sense, ovens around supper time and dishwasher later. Probably laundry more on weekends also
